weighted-centroid-distance 1.0.0

Python serverside toolkit

Weighted Centroid Distance

A new model assessment technique for multi-class classification tasks in machine learning.

For the statistical theory, please see the white paper in /paper. It is auto generated from .tex to .pdf.

How to use

# Python
from weighted_centroid_distance import WeightedCentroidDistance
distribution = WeightedCentroidDistance.get_distribution([1, 2, 3, 3, 4, 5, 5, 5], inverse=False)
wcd = WeightedCentroidDistance(distribution=distribution)
y____ = [1, 1, 2, 2, 3, 3, 3, 4, 4, 5, 5, 5, 5, 5, 5]
y_hat = [1, 1, 2, 2, 3, 3, 4, 4, 4, 4, 5, 1, 5, 5, 2]
res = wcd.get_distance(y=y____, y_hat=y_hat)
print("res: ", res)
# 0.12086457016125143

Todo

• Create graphs for all parts of the algorithm defined in the paper
• Improve python script, and make a versions for other languages.